Within hours of DeepSeek releasing its V4 model preview on April 24, 2026, Chinese technology companies began moving to lock down supplies of Huawei’s domestically built AI processors, according to sources familiar with the procurement activity. The rush reflects a market that has been quietly preparing for this moment: a homegrown AI model powerful enough to justify large-scale investment in Chinese-made chips, arriving at a time when U.S. export controls have made the alternative nearly impossible to obtain.
The V4 release, reported by the Associated Press, runs on Huawei’s Ascend-series accelerators, the most prominent being the Ascend 910B and its successor, the 910C. These chips, manufactured by Semiconductor Manufacturing International Corporation (SMIC) using older deep-ultraviolet lithography, represent China’s best current answer to Nvidia’s H100 and H200 processors, which Beijing’s firms can no longer legally import after successive rounds of U.S. export restrictions tightened in October 2022, October 2023, and subsequent updates through 2025.
Same-day enterprise adoption signals commercial confidence
The clearest public evidence of V4’s commercial pull came the same day the preview went live. Aurora Mobile announced that its GPTBots.ai platform had integrated DeepSeek V4 preview, offering enterprise users million-token context windows and what the company described as next-generation agentic AI capabilities.
That timeline matters. Same-day integration suggests Aurora Mobile either received early access to V4 or had engineers standing by to move the moment the model dropped. Either scenario points to strong commercial conviction in the underlying technology. The company’s platform serves enterprise workflows that depend on processing large volumes of unstructured data, including legal review, compliance checks, and customer support analytics.
The million-token context window is more than a spec-sheet number. It means an enterprise application can ingest an entire case archive, a full codebase, or years of transaction records in a single pass, without splitting the data into smaller chunks that risk losing coherence. For legal teams, financial institutions, and software development shops, that capability opens workflows that were previously impractical. Aurora Mobile is treating V4 not as a standalone product but as an infrastructure upgrade embedded into its existing platform, quietly raising the ceiling on what its customers can automate.
The procurement picture remains incomplete
The core claim that multiple Chinese firms are scrambling for Huawei chips rests on unnamed sources, and important details have not surfaced publicly. No specific procurement volumes, contract values, or delivery timelines have been disclosed by Huawei, DeepSeek, or any buyer beyond Aurora Mobile. Whether these orders represent a sharp spike triggered by V4 or incremental additions to contracts already in motion is unclear.
There is also a supply question. Huawei’s Ascend chips depend on SMIC’s fabrication lines, which use older lithography and have faced scrutiny over yield rates and throughput. Industry watchers have questioned whether SMIC can scale production fast enough to meet a demand surge without running into allocation bottlenecks or forcing buyers to accept longer lead times and higher prices. If V4 drives the kind of broad adoption its early reception suggests, Huawei’s ability to deliver chips at volume becomes a bottleneck worth watching.
Performance comparisons add another open question. DeepSeek has not published detailed benchmarks showing how V4 performs on Huawei Ascend chips versus other hardware. Without that data, it is hard to tell whether V4 is specifically optimized for Huawei’s architecture or whether Chinese firms are choosing Ascend processors primarily because Nvidia’s latest chips are off the table. The distinction matters: optimization would signal a structural shift toward domestic hardware, while necessity alone would suggest the arrangement is more fragile and could reverse if export controls ever loosen.
Aurora Mobile’s announcement, while concrete, describes a software deployment, not a hardware purchase. The company did not disclose chip sourcing, data center locations, or infrastructure spending. It could be running V4 on Huawei Cloud resources it does not own, consuming the model as a managed service. Treating the integration as direct proof of a broader hardware rush requires a logical step the public documentation does not fully support.
Why this moment matters for the global chip divide
Even with gaps in the procurement data, the V4 release lands at a structurally significant moment. China’s AI sector has spent the past three years adapting to life without cutting-edge Nvidia silicon. Companies like Baidu, Alibaba, and Tencent have invested in Huawei’s cloud AI services, while smaller players have explored chips from domestic startups like Biren Technology and Cambricon. But none of those alternatives had a flagship model release to rally around, a proof point showing that competitive AI could be built and run entirely on Chinese hardware.
DeepSeek’s progression tells part of that story. Its V3 model, released in late 2024, drew global attention for achieving strong benchmark results at a fraction of the training cost of comparable Western models. The R1 reasoning model followed, pushing further into complex problem-solving tasks. V4 appears to extend that trajectory with its long-context capabilities and agentic features, though independent benchmarks have not yet confirmed how it stacks up against the latest offerings from OpenAI, Anthropic, or Google DeepMind.
For U.S. chipmakers, the implications are tangible. Nvidia has already disclosed that export controls cost it billions in potential China revenue. If V4 accelerates a shift toward Huawei-based AI infrastructure, that lost revenue becomes harder to recover even if restrictions are eventually relaxed. Customers who build their software stacks around Ascend chips and DeepSeek models will face switching costs that discourage a return to Nvidia, much the way enterprises that migrated to cloud platforms rarely move back to on-premise servers.
For businesses and investors watching from outside China, the key question is whether V4 marks the beginning of a self-sustaining cycle: a competitive model drives chip demand, chip demand funds expanded production, expanded production enables the next generation of models. That cycle has powered Nvidia’s dominance in the West. Whether Huawei and DeepSeek can replicate it depends on factors that remain unresolved, including fabrication yields, software ecosystem maturity, and the willingness of China’s private sector to commit capital at scale.
What to watch in the months ahead
Several concrete indicators should clarify the picture before mid-2026. Quarterly earnings from Chinese cloud providers will reveal whether capital expenditure on AI infrastructure is accelerating. Huawei’s own disclosures, or leaks from its supply chain, could confirm whether Ascend chip production is ramping. Additional enterprise announcements referencing V4 deployments would broaden the evidence beyond Aurora Mobile’s single data point. And independent benchmark results, if DeepSeek or third parties publish them, would answer the optimization question that currently hangs over the hardware story.
What can be said with confidence today is narrower but still significant. DeepSeek’s V4 has created a new demand signal for domestic AI infrastructure in China. Its capabilities raise the computational bar for any firm competing in that market. And for companies that cannot access Nvidia’s latest processors, the incentive to build on Huawei’s stack just became considerably stronger. The scale of the resulting chip rush, though, remains a question the available evidence cannot yet answer.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
